InPO: Inversion Preference Optimization with Reparametrized DDIM for Efficient Diffusion Model Alignment